The Design Inference

Dembski's explanatory filter removes only the chance hypotheses fed to it. Its one worked biological number prices random assembly, which no biologist proposes, and its author set it aside. A specification may be found after the fact. The test that survives uses a chance model with selection; it has been computed for a few mutations, never for a molecular machine.

By AI (Claude Fable 5.1, reasoning effort 30); strengthening review by Claude Opus 5.5, thinking level unavailable Draft:  September 2026

"before we even begin to send E through the Explanatory Filter, we need to know what probability distribution(s), if any, were operating to produce the event." — William Dembski, The Design Inference, 19981

"The example in NFL ch. 5 doesn't fit the bill." — William Dembski, on his own flagellum calculation, 20082

Summary

William Dembski's design inference is a rule for rejecting chance.  If an event matches an independent pattern and is improbable enough under every relevant chance hypothesis, the event is assigned to design.  The logic is sound.  The trouble is in the inputs, and Dembski has said so himself at each stage.

  • The filter removes only what is put into it. Dembski wrote in 1998 that the user must already know which probability distributions were operating.1 In 2005 he wrote that the list of chance hypotheses may always omit "some crucial chance hypothesis."3 Branden Fitelson, Christopher Stephens and Elliott Sober made the same point against him in 1999.4
  • The one worked biological example prices a hypothesis nobody holds. The flagellum figure in No Free Lunch, about 1 in 10234, is the chance of pulling 250 suitable proteins at random from the 4,289 that E. coli makes.5 Evolutionary biologists propose staged co-option under selection.6 Dembski wrote in 2005 that the probability his argument needs "has yet to be done," and in 2008 that the example "doesn't fit the bill," because a test case must allow a probability calculation that can be "convincingly performed." He pointed instead to Douglas Axe's estimate for a single protein fold.32
  • A specification does not have to come before the event. The catalogue description of this article said it did. Dembski says the opposite, and two logicians who never set out to defend him reach the same result.17 What must be fixed first is the rule for what counts as a match.
  • "No false positives" cannot be tested on natural objects as Dembski frames it. He names the Oklo natural reactor as the best candidate for a false positive, then says that if Oklo passed the test it would show that Oklo was designed.55 He does grant that any single verdict can be mistaken, and the 2013 version of his measure states a false-positive rate under a named chance model.58
  • Ordinary design detection names its hypotheses. Fisher on Mendel, forensic laboratories, election forensics, the first pulsar and the eolith dispute each state a chance model, and most state what the agent would have done.
  • Durston's measure is a measurement, not an inference. His 2007 paper counts functional bits against an equal-probability baseline.9
  • Verdict. The defensible claim is narrow: a stated pattern, a stated chance model that includes selection, and a computed probability. Design researchers and their critics have both done that calculation for changes of a few mutations. Nobody has done it for a molecular machine.

Part 1: The method in Dembski's own words

The Design Inference was published by Cambridge University Press in 1998.  It asks one question of any event:  "are we going to attribute it to regularity or chance or design?"1  The explanatory filter is a flowchart with three tests in order.  A highly probable event goes to regularity.  An event of intermediate probability goes to chance.  An event of small probability goes to design only if it is also specified.

Four features of the 1998 book matter for everything that follows.

Design is defined as a leftover.  Dembski calls design "the complement of regularity and chance" and says the advantage is that this "avoids committing itself to a doctrine of intelligent agency."1  He is explicit that "design as inferred from the design inference does not logically entail an intelligent agent."1  So the filter, as first stated, separates events that chance can be blamed for from events it cannot.  It does not identify a designer.

All relevant chance hypotheses must go.  "In eliminating chance, the design inference eliminates not just a single chance hypothesis, but all relevant chance hypotheses."1  Dembski contrasts this with ordinary statistics, where rejecting one distribution leaves others open.  The price is stated in the pull quote above:  the user must know beforehand which distributions were operating.

Specification is independence, not priority in time.  The running example is Nicholas Caputo, a New Jersey county clerk who gave Democrats the top ballot line in 40 of 41 drawings.  Nobody wrote that pattern down in advance.  Dembski's comment:  "Though discovered after the fact, this pattern is not a fabrication."  Patterns given beforehand are only "a proper subset of the patterns that legitimately eliminate chance," and "the important thing about a pattern is not when it was identified, but whether in a certain well-defined sense it is independent of an event."1  He calls this detachability and tests it with a question:  "Given an event, would we be able to formulate a pattern describing it if we had no knowledge which event occurred?"1

The universal probability bound counts specifications, not trials.  Dembski multiplies an upper limit of 1080 elementary particles by 1045 state changes per second and by 1025 seconds, a span about a billion times the age of the universe.1  The product, 10150, caps how many patterns any observers could ever have specified.  The bound is half its reciprocal, "1/2 x 1/10150."1  The common wording, "1 in 10150," drops the factor of one half.

Part 2: How the method changed, 2002 to 2023

2002.  No Free Lunch renames the criterion complex specified information and sets the threshold at 500 bits, the bound restated in bits.5  It adds a Law of Conservation of Information whose conclusion is that "natural causes are incapable of generating CSI."5  The 1998 book kept design apart from agency.  The 2002 book claims the criterion reliably detects intelligence.

2005.  "Specification:  the pattern that signifies intelligence" replaces detachability with descriptive complexity.  A pattern is a good specification when it can be described briefly.  The measure is χ = –log2[10120 · φS(T) · P(T|H)], where φ counts the patterns at least as simple to describe as T, P(T|H) is the probability of the pattern under chance hypothesis H, and 10120 is Seth Lloyd's bound on operations in the universe.3  A pattern counts as a specification when χ is greater than 1.3  Since 10120 is about 2399, the 2005 threshold is about 400 bits plus the bits needed to count the rival patterns, against 500 bits in 2002.10  The measure is always relative to a named H.

2008.  On 3 December, in a comment thread about Peter Olofsson's criticisms, Dembski wrote:  "I've pretty much dispensed with the EF. It suggests that chance, necessity, and design are mutually exclusive. They are not."2  Seven days later he posted a retraction:  "I'm herewith reinstating it — it will appear, without reservation or hesitation, in all my future work on design detection."11  The retraction covers the filter.  It does not mention the comment's second point, on the flagellum, taken up in Part 4.

2013 to 2015.  Winston Ewert, Dembski and Robert Marks define algorithmic specified complexity as ASC(X, C, P) = −log2 P(X) − K(X|C):  the improbability of X under distribution P, less the length of the shortest program that produces X given context C.8  The chance of seeing at least α bits is at most 2−α.  They state the scope plainly:  "ASC provides evidence that a stochastic outcome modeled by the distribution, P(X), does not explain a given object."12  That is a claim about one distribution.

2023.  The second edition, with Ewert as co-author, was not obtained in full.  Its contents pages, two publisher's excerpts, Dembski's own summary and Joe Felsenstein's section-by-section report were read.  The filter is kept but moves to section 5.6.  New chapters cover specified complexity by description length, with frequentist and Bayesian readings, and evolutionary biology.13  Dembski's summary:  "we define specifications as patterns with short descriptions," and specified complexity is "SC ( E ) = I (E) – K ( E )," the same quantity as ASC.1414  The Introduction now rests the biological case on conservation of information:  "To seal the deal, however, the ID research community still needed something like the second law for biology."15  That argument is examined in Evolutionary Computation.

Part 3: How uncontroversial design detection works

Dembski holds that forensic science, cryptography, archaeology and the search for extraterrestrial intelligence already use his logic.  Seven cases were checked.

Field and case Chance model stated Design hypothesis stated How it was decided
Statistics: Fisher on Mendel, 1936 Random sampling around Mendel's ratios An assistant "who knew too well what was expected" One model rejected, then a named agent with a motive16
Forensic science: European guideline, 2015 One party's account The other party's account Likelihood ratio between a pair of propositions17
Election forensics: Russia, 2011 to 2012 Vote distribution of normal elections Ballot stuffing, with fitted parameters A model of the fraud itself18
Radio astronomy: the first pulsar, 1967 Unknown natural source A transmitter on a planet What a planet would do to the signal19
Cryptology: Shakespeare ciphers, 1957 Letters falling by accident under fixed rules A person enciphering with a stated key One model rejected, rules fixed first20
Archaeology: eoliths, 1905 Natural fracture, measured by experiment A human knapper Comparison with what nature produces21
Nuclear safeguards: Oklo, 1972 Natural ore at 0.7202% uranium-235 Uranium already used in a plant A named natural process, confirmed22

Three things stand out.

First, elimination is normal practice.  Ronald Fisher tested Mendel's data against one model and found a deviation "only to be expected once in 444 trials."16  William and Elizebeth Friedman, professional cryptologists, wrote that "the chances calculated exactly" settle whether a cipher key is real.20  Dembski is right that chance can be rejected without first pricing design.  Sober's strictest statement, that probabilistic rejection "is an incorrect form of inference," would condemn Fisher's paper too.23

Second, elimination works in these cases because the chance model is known.  A fair draw or a binomial sample is a single distribution.  The European forensic guideline states the limit:  "If no alternative can be formulated, the value of the findings cannot be assessed."17

Third, the design hypothesis usually predicts something.  Jocelyn Bell Burnell's group reasoned that pulses from a planet "should show Doppler shifts as the little green men on their planet orbited their sun."  The only shift found was the earth's own.  A second source then appeared elsewhere in the sky, and "it was very unlikely that two lots of little green men would both choose the same, improbable frequency."1919  Sober makes the point about SETI in general:  engineers look for narrow-band signals "simply because they think they know what sorts of mechanisms are needed to produce it," and not because the signal meets a prior test of complexity.24

SETI's own documents use both kinds of reasoning.  The Breakthrough Listen team searches for narrowband signals because such a signal "is too narrow to arise naturally from known natural astrophysical sources, and represents a power-efficient method of transmitting a beacon signal out to great distances."25  The first half is Dembski's elimination of known natural causes.  The second half is Sober's model of what a transmitter would do.  The International Academy of Astronautics' protocol of 1989 asks a discoverer to verify "that the most plausible explanation for the evidence is the existence of extraterrestrial intelligence rather than some other natural phenomenon or anthropogenic phenomenon."  If that cannot be confirmed, the discoverer may report the signal "as appropriate to the discovery of any unknown phenomenon."26  So SETI eliminates natural and human causes, compares explanations, and keeps an outcome for "unknown."  The last of these matters in Part 6.

Part 4: The flagellum calculation

Chapter 5 of No Free Lunch holds the only probability Dembski has computed for a biological structure.  He treats the flagellum as a "discrete combinatorial object" whose probability is "the product of these three probabilities":  originating the parts, localizing them, and configuring them.5

Factor Dembski's model His figure
Localization Draw 250 proteins at random from E. coli's 4,289, needing all to fall among 500 suitable ones (500/4,289)250, about 10−2345
Origination Each 300-residue protein treated as a random string with 10% tolerance About 10−39 per protein5
Configuration Approximated by a perturbation probability Far below the bound

The arithmetic is right.27  The model is Dembski's own analogy:  a shopper filling a cart at random in a supermarket.

What hypothesis does this eliminate?  It eliminates assembly by uniform random draw.  Peter Olofsson, a probabilist, put it this way in 2008:  Dembski "only tests one particular chance hypothesis and takes its rejection as evidence that all chance hypotheses can be ruled out," and that hypothesis is "that the flagellum is assembled at random."28  Olle Häggström had already noted that uniform distributions "have no privileged status over other models as realistic descriptions of what the laws of nature produce."29

Does any biologist propose it?  No.  Nicholas Matzke's 2003 model has "six major stages: export apparatus, secretion system, adhesion system, pilus, undirected motility, and taxis-enabled motility," each argued to be useful on its own.6  Renyi Liu and Howard Ochman argued in 2007 that the core flagellar genes "were derived from one another" by duplication.30  Other biologists criticised that paper's homology claims.  It is cited here only for the kind of hypothesis on offer.

Dembski's position, stated fairly.  He does not claim biologists believe in random assembly.  His argument has two steps.  Michael Behe's irreducible complexity is taken to show that such systems "must be attained all at once."5  With selection removed, random assembly is what is left to price.  Jason Rosenhouse objects that the calculation "is completely parasitic on Michael Behe's claims":  if Behe is right, the number adds nothing.31  Dembski's 2023 Introduction describes the same relation:  his approach was "to piggyback on the work of design biologists."15

Dembski's own later assessments.  In 2002 he called it a challenge "to firm up these numbers and make sure they do not cheat in anybody's favor."5  In 2005 he defined the hypothesis properly, as "an evolutionary chance hypothesis that takes into account Darwinian and other material mechanisms," and then wrote:  "The precise calculation of P(T|H) has yet to be done."3  In 2008 he wrote that a test case must be "simple enough on which the probability calculation can be convincingly performed but complex enough so that it does indeed exhibit CSI."  The flagellum "doesn't fit the bill," and Axe's protein-fold example "is much stronger."2  So he set the flagellum aside because its probability could not be convincingly computed, not because he thought it arose by chance.  According to Felsenstein, the 2023 edition's worked example is Douglas Axe's estimate for a beta-lactamase fold, not the flagellum.32  That estimate is examined in How Rare Is a Working Protein?, which finds it a conditional upper bound sixty-six orders of magnitude from the rival measurement.

Dembski's answer when the number cannot be computed.  The Design Revolution (2004) argues that a design claim can be justified without P(T|H).  Dembski's word is assertibility:  whether a claim is rationally justified by the evidence in hand.  He writes that "science must form its conclusions on the basis of available evidence, not on the possibility of future evidence."  If biologists produce "detailed, testable, indirect Darwinian pathways" for the flagellum, "intelligent design will quickly pass into oblivion."33  He grants "the logical impossibility of ruling out the infinity of possible indirect Darwinian pathways" and still calls the flagellum's specified complexity "highly assertible."33  He adds an in-principle argument, which he credits to Michael Polanyi and Stephen Meyer.  The space of possible DNA or protein sequences is "perfectly homogeneous," so its geometry "precludes any underlying mechanisms from distinguishing or preferring some character strings over others."33  This is the design side's best case for using the inference without the number.  Its standard is fair:  science judges on the evidence it has, and biologists have no computed pathway either (Objection 2).  It has two limits.  It is a judgement held "at least provisionally," so it no longer rests on the filter's calculation or the universal bound.  And the homogeneity premise rules out chemical preference among sequences.  Selection is not a chemical preference:  it favours sequences by what they do.  Dembski's answer to that, conservation of information, is examined in Evolutionary Computation.

What the flagellum argument claims without a number is examined in Irreducible Complexity.

This is the crux.  The filter's conclusion is only as good as the list of chance hypotheses, and for the flagellum the list has one entry that no opponent defends.

Part 5: The specification problem

The Texas sharpshooter fires at a barn and paints the target around the holes.  Any outcome can be given a pattern afterwards.  The question is what stops this.

Dembski's first answer, in 1998, was detachability:  the pattern must be one a person could have produced without seeing the event.  His second, from 2005, is brevity:  the pattern must have a short description.  Yuri Gurevich and Grant Passmore, logicians working on lottery fraud, reached the same idea independently.  Their conclusion is that "the point is not that the focal event is specified in advance but that it is specified — possibly after the fact — without any knowledge about the outcome of the trial."7  On this point Dembski is sound.

Two problems remain.

The rule must be fixed before the probability is computed.  The Friedmans examined the ciphers claimed in Shakespeare, each a pattern found after the fact.  Their test was whether the rules are stated so exactly that "the chances of its appearing by accident" can be calculated.20  A specification is only as good as the count of other patterns the same looseness would have accepted.  Dembski's φ is meant to be that count.

For the flagellum the count was never made.  Dembski wrote in 2002:  "Specification is never a problem," because "humans developed outboard rotary motors well before they figured out that the flagellum was such a machine."5  The concept is independent of the event, and that is not disputed here.  But the probability that goes with a pattern is the probability of everything that fits the pattern.  Olofsson's first objection is that Dembski "considers only the outcome, not the rejection region," and asks how one could list everything with "at least the function of the flagellum."28  In 2005 Dembski gave the pattern as "bidirectional rotary motor-driven propeller" and estimated 1020 descriptions at least as simple.3  That number is a vocabulary count.  The share of protein arrangements that would make some rotary propeller is unknown.

So the pattern is fixed independently of the fact.  Its size as a target is not fixed at all.  The 2023 Introduction still treats the matter as closed:  "the specification of these systems ... was never in question."15

Part 6: False positives

Dembski grants false negatives freely, since a designer can imitate chance.  On false positives his claim is strong.  The criterion is a net, and "if things end up in the net that are not designed, the criterion is vitiated."5  His claim is that it "successfully avoids false positives."5  The argument is inductive:  wherever the causal story is known, design is present.

He does not claim that any single verdict is final.  In 1998 he wrote that eliminating chance "is never a logical deduction," and that "the possibility always remains that it was a mistake to eliminate chance."1  In 2002 he allowed that a user may "later discover that we missed a crucial" chance hypothesis, in which case the inference "would be mistaken."  Such errors are "the risks of empirical inquiry, which of its nature is fallible."5  What he claims is that the criterion does not err when it is given the right chance hypotheses.  The dispute is over whether the right hypotheses can be known for living things.

Case What happened Does it count against the claim?
Kepler's lunar "towns," 1623 Kepler inferred from the circular shape of craters that moon-dwellers dug them for shade, with a stake and rope34 Partly. Shape plus function, no mechanism known. Kepler called the remarks "playful"34
Eoliths, 1905 Chipped flints were accepted as tools by many geologists and archaeologists. Hazzledine Warren argued from experiment that natural pressure makes them21 Yes, for archaeology's inference. Wild macaques now produce flakes that "would likely be identified as anthropogenic"35
LGM-1, 1967 Regular radio pulses. The discoverers "had no proof that it was an entirely natural radio emission"19 No. A periodic signal falls to regularity. But no regularity was known when design was dropped
Snowflakes Six arms grow alike because "all six arms experience the same changing conditions"36 No. Dembski answers that a snowflake has one kind of part5
Oklo, 1972 Uranium-235 was low. The plant's routine check was for uranium already used in a reactor. Fission products in the ore showed a natural chain reaction22 Undecided, and the way it is undecided is the problem

Dembski raises Oklo himself as "perhaps the best candidate" for a false positive.5  He says the probabilities "have yet to be ascertained" and expects them not to be small.  Then:  "But suppose the Oklo reactors ended up satisfying this criterion after all. Would this vitiate the complexity-specification criterion? Not at all. At worst it would indicate that certain naturally occurring events or objects that we initially expected to involve no design actually do involve design."5

With that move, no natural object can count as a false positive.  If it passes the test, it is reclassified as designed.  Wesley Elsberry and Jeffrey Shallit object to the other half of the induction:  by demanding "video-camera-certainty," it "rules out considering cases where the underlying causal story cannot be known."37  The induction therefore covers human artifacts only, where nobody disputes design.

The later formalism treats false positives the way a significance test does.  Under a named distribution P, the chance that an object scores at least α bits of algorithmic specified complexity is at most 2−α.8  At 500 bits a false alarm under P has a probability of at most 2−500.  Ewert, Marks and Dembski draw a modest conclusion from a high score:  "we have evidence that the assumed probability distribution was incorrect."8  A stated, tiny error rate is firmer ground than "no false positives," and it is the right form for the claim.  It does not cover the Oklo kind of error.  A natural process nobody has modelled is a distribution nobody wrote down, and the bound says nothing about distributions left off the list.

None of the historical cases is a clean failure of Dembski's criterion, because in none was a probability computed against a bound.  They are failures of the informal inference the criterion formalises.  They show what John Wilkins and Elsberry said the filter lacks:  "a 'Don't know' decision."38  The pulsar team used one.

Dembski answered them by name in 2002.  The risk of leaving out a natural cause is "a danger endemic to all of scientific inquiry," "merely a restatement of the problem of induction."5  Elsberry's added outcome, "unknown causation," he called "a disingenuous display of ignorance," because "intelligence and only intelligence is known to generate specified complexity."5  The first half of the reply is fair:  every empirical method can miss a cause.  The second half rests on the premise in dispute, and SETI's protocol keeps the very outcome Dembski rejects (Part 3).

Part 7: The critics and the replies

Critic Central point, with its reason Reply
Fitelson, Stephens and Sober, 1999 The filter "applies only to agents who believe they have a complete list of the chance processes." Paley compared two specific hypotheses and needed no such list44 Dembski 2002: "Why should eliminating a chance hypothesis always require additional chance hypotheses that compete with it?"5
Sober, 2003 and 2008 Design confers a probability only given "the designer's goals and abilities"24 "We infer design regularly and reliably without knowing characteristics of the designer."5 Later: Bayesian method "is parasitic on the Fisherian approach"39
Wilkins and Elsberry, 2001 No "don't know" outcome, so ignorance is scored as design Dembski 2002: a missed cause is "a danger endemic to all of scientific inquiry," and an "unknown causation" outcome is "a disingenuous display of ignorance"55
Wolpert, 2002 The No Free Lunch argument is "fatally informal and imprecise"40 Later papers with Marks restate the argument formally
Häggström, 2007 The uniform distribution over fitness landscapes is unrealistic Later work drops that use of the theorems
Felsenstein, 2007 The conservation proof works only "by changing the specification"41 None found
Olofsson, 2008 One uniform hypothesis, and no rejection region, for the flagellum Dembski's comment of 3 December 2008 concedes the example
Elsberry and Shallit, 2011 An event with "obscure causal history could mistakenly be assigned to design"37 No formal reply. The 2005 redefinition came after their 2003 draft
Devine, 2014 "Natural explanations are eliminated before chance," so the order of the filter favours design42 None found
Felsenstein, 2024 The second edition gives no "tail probability in some relevant distribution"32 None found

Wolpert, Häggström and the conservation argument are treated in Evolutionary Computation and are not re-argued here.  "None found" means a search of Dembski's site, the Evolutionary Informatics Lab publication list and the Discovery Institute's news site found no reply.

Sober concedes more than is usually noticed.  Archaeologists, he writes, "often have objects in their collections that they know are artifacts, although they have no idea what the makers of these artifacts had in mind."24  The 1999 notice calls the filter "much too hard on Design."4  The strongest critics reject the filter.  They do not say design cannot be inferred.

Part 8: Durston's measure, and what survives

Kirk Durston, David Chiu, David Abel and Jack Trevors published a method in 2007 for counting functional bits, or Fits, in a protein family.  The baseline is a null state with "the equi-probability of all possible sequences."9  The paper draws no design inference.  It says function "could theoretically arise stochastically through mutational changes coupled with selection pressure, or through human experimenter involvement."9

Durston's design argument is in his popular writing, and it is inductive:  "A unique property of intelligent minds is the ability to produce statistically significant levels of functional information."43  He says one verified natural process would falsify it.  This is a different kind of argument from Dembski's.  It claims that only minds have been seen to do this.  It does not claim that chance has been excluded.

As a specification measure, a Fit count supplies the −log2 P term under one distribution, the uniform one.  It has the same limit as the flagellum figure.  The numbers, and the test in which unconstrained simulated families scored like real ones, are in Measuring Biological Information.  Felsenstein reports that the 2023 edition has moved away from functional information to description length.44

The claim that survives has four parts:

  1. A pattern T stated by a rule fixed independently of the object, with a count of the rival patterns the rule would also accept.
  2. A named chance hypothesis H that includes mutation, selection, duplication and co-option. A uniform draw does not qualify.
  3. A computed P(T|H), with the number of trials stated.
  4. A conclusion no wider than this: H does not explain T. That is what Ewert, Dembski and Marks claim for ASC.

Gurevich and Passmore endorse this form for a lottery and call the wider ambition, to "impugn any chance whatsoever," hopeless.7

The evidence needed for part 2 is the fraction of sequences that function, how far apart they sit, the measured rate at which populations cross such gaps, and the population sizes and generations available.  A waiting-time calculation under selection is a P(T|H) of the right kind, and both sides have published them for small targets.  Michael Behe and the physicist David Snoke modelled a duplicated gene that needs two or more changes, none useful alone.  Fixing such a feature in 108 generations "would entail population sizes of no less than 109."45  Michael Lynch replied that with neutral intermediates such features evolve "on time scales of a million years (or much less)."46  Rick Durrett and Deena Schmidt, writing against Behe, found a few million years enough for one named pair of changes in fruit flies, but in the human line the same change "would take >100 million years."47  Ann Gauger and Douglas Axe estimated "some 1030 or more generations" for an E. coli enzyme to take over its neighbour's job after duplication.  They concluded only that "there is now a scientific case for doubting this particular jump."48  John Sanford and colleagues simulated a human-ancestor population in which a named string of two DNA letters "required on average 84 million years" to become fixed.49

These are the filter's missing term, computed for the smallest cases.  The numbers disagree mainly over two inputs:  whether the intermediate steps help or harm, and how many different changes would do the job.  Most of the calculations price one named change.  Durrett and Schmidt note that if a new binding site may arise anywhere in a long stretch of DNA, "then there are many more chances."47  That is the rejection-region problem of Part 5 again.  The experiments and the replies are in Gene Duplication and Divergence and Benchmarking Evolution.  None of these calculations covers a molecular machine, and none was run against the universal bound.

Objections and responses

1.  "Requiring every chance hypothesis to be eliminated would make any design inference impossible, including a murder conviction."  This is the design side's best reply, and it is correct.  Dembski does not claim a verdict is final.  He calls the inference fallible:  it "would be mistaken" if a missed chance hypothesis turned up (Part 6).5  The difference from a court lies elsewhere.  A court knows the rival accounts of a death.  For the origin of a molecular machine, the list of chance hypotheses is the thing in dispute.

2.  "Biologists have no detailed pathway for the flagellum either, so they have no probability to offer."  True.  Matzke's six stages are a sketch, not a computation.  P(flagellum | evolution) is unknown.  An unknown probability cannot be entered in the filter as a small one.

3.  "Function is an objective specification."  Function is independent of the observer, which answers the sharpshooter charge.  It does not supply the size of the target.

4.  "If the method fails, design in biology is refuted."  It is not.  A failed detector shows nothing about what is there.  Leslie Orgel wrote in 1973 that "living organisms are distinguished by their specified complexity," and Richard Dawkins in 1986 that they have "some quality, specifiable in advance, that is highly unlikely to have been acquired by random chance alone."5051  The observation is common ground, as Organisms Appear Designed documents.  The dispute is over what explains it.

Issues and unanswered questions

Against the design side.

  • Probabilities under hypotheses that include selection have been published only for changes of two to about seven mutations (Part 8). None has been published for a molecular machine, and none has been run against the universal bound. After twenty-eight years this is the main gap.
  • The flagellum calculation was set aside by its author in 2008. This site should not use it.
  • The "no false positives" claim is insulated from test by the Oklo passage. Either a natural object with high specified complexity refutes the criterion, or the criterion cannot be refuted by natural objects. Dembski chose the second.
  • Devine's point, that the order of the filter's tests favours design, has no published answer.
  • The second edition was not read in full. If its section 7.6 prices a hypothesis that includes selection, this article must change.

Against the critics.

  • The likelihood approach needs the probability of a flagellum given a designer. Sober admits this depends on unknown goals and abilities. The critics' preferred method cannot be run either.
  • Sober's rejection of all probabilistic elimination proves too much. Fisher, the Friedmans and every significance test use it.
  • "An unknown pathway may exist" is unfalsifiable if it is allowed to stand indefinitely. Dembski's complaint that critics "merely needed to postulate unidentified" pathways is fair.15
  • The reviews of the second edition are blog posts, and Rosenhouse wrote his before reading the book.

Open for both.

  • Can a rejection region be defined for a molecular machine?
  • The fine-tuning argument has the same structure. The origin of the first replicator, under abiogenesis, is the one biological case where selection is not available to H.

Verdict

The design inference is a correct rule for rejecting a stated chance hypothesis.  Cryptology and fraud detection use that rule, and Dembski's account of after-the-fact specification is sound.  It is not yet a method for telling designed organisms from evolved ones.  The only biological probability computed with it prices random assembly, its author has withdrawn that example because its probability cannot be convincingly computed, and the claim of no false positives is framed so that no natural object could count against it.  The repair is known and both sides have named it:  state the pattern and its rivals, state a chance hypothesis with selection in it, and compute.  Both sides have done this for changes of a few mutations.  Until someone does it for a molecular machine, this site should present specified complexity as an accurate description of living things and not as a proof.

Sources

  1. 1.

    Dembski, William A.  The Design Inference: Eliminating Chance through Small Probabilities.  Cambridge University Press.  1998.  Dembski's first book on the method, a monograph in probability and philosophy of science.  It sets out the explanatory filter, specification and the universal probability bound, defines design as whatever remains once regularity and chance are eliminated, and states the method's limits in its own terms.

    1. 1a.

      Page 8:  “In eliminating chance, the design inference eliminates not just a single chance hypothesis, but all relevant chance hypotheses.” ↩

    2. 1b.

      Page 9:  “design as inferred from the design inference does not logically entail an intelligent agent.” ↩

    3. 1c.

      Page 14:  “Though discovered after the fact, this pattern is not a fabrication. Patterns given prior to an event, or Peirce's predesignations, constitute but a proper subset of the patterns that legitimately eliminate chance. The important thing about a pattern is not when it was identified, but whether in a certain well-defined sense it is independent of an event.” ↩ ↩

    4. 1d.

      Page 15:  “Given an event, would we be able to formulate a pattern describing it if we had no knowledge which event occurred?” ↩

    5. 1e.

      Page 36:  “are we going to attribute it to regularity or chance or design?” ↩

    6. 1f.

      Page 36:  “The principal advantage of characterizing design as the complement of regularity and chance is that it avoids committing itself to a doctrine of intelligent agency.” ↩

    7. 1g.

      Page 41:  “before we even begin to send E through the Explanatory Filter, we need to know what probability distribution(s), if any, were operating to produce the event.” ↩ ↩

    8. 1h.

      Page 209:  “within the known physical universe there are estimated to be no more than 10^80 elementary particles. Moreover, the properties of matter are such that transitions from one physical state to another cannot occur at a rate faster than 10^45 times per second. Finally, the universe itself is about a billion times younger than 10^25 seconds”.  The OCR prints each exponent as a degree sign.  The three are 80, 45 and 25:  they must sum to 150, the age clause fixes 25, and No Free Lunch p. 156 restates the product as 1 in 10^150, or 500 bits. ↩

    9. 1i.

      Page 214:  “In justifying the universal probability bound”.  The sentence continues with the bound in symbols, one half times 1 over 10^150.  The OCR garbles the symbols.  Note 18 on p. 213 gives Borel's bound and calls Dembski's a rigorous version of it. ↩

    10. 1j.

      Page 192:  “the elimination of chance in (10) is never a logical deduction”.  “the possibility always remains that it was a mistake to eliminate chance.”  Steps (1) to (10) are the numbered premises and conclusion of Dembski's formal argument for eliminating chance.  The exception he names is an event of probability zero. ↩

  2. 2.

    Dembski, William A.  “Comment on "Some Thanks for Professor Olofsson".”  Uncommon Descent.  Dec 3, 2008.  Posted 6:24 PM PDT under a post by DaveScot of 25 November 2008.  On the archived page each author's name follows the comment.

    1. 2a.

      Paragraph point 1:  “I've pretty much dispensed with the EF. It suggests that chance, necessity, and design are mutually exclusive. They are not. Straight CSI is clearer as a criterion for design detection.” ↩

    2. 2b.

      Paragraph point 2:  “The challenge for determining whether a biological structure exhibits CSI is to find one that's simple enough on which the probability calculation can be convincingly performed but complex enough so that it does indeed exhibit CSI. The example in NFL ch. 5 doesn't fit the bill. The example from Doug Axe in ch. 7 of THE DESIGN OF LIFE (www.thedesignoflife.net) is much stronger.”  The book named is Dembski and Jonathan Wells, The Design of Life, 2008. ↩ ↩ ↩

  3. 3.

    Dembski, William A.  Specification: The Pattern That Signifies Intelligence.  Aug 15, 2005.  Version 1.22 of the author's manuscript, 41 pages, numbered at the foot.  Page numbers here are the manuscript's.  Later printed in Philosophia Christi 7(2):299–343.

    1. 3a.

      Page 18:  “the molecular machine known as the bacterial flagellum can be characterized as a 4-level concept of the form “bidirectional rotary motor-driven propeller.”” ↩

    2. 3b.

      Page 24:  “We thus define the specified complexity of T given H (minus the tilde and context sensitivity) as”.  The formula follows, with 10 to the power 120. ↩

    3. 3c.

      Page 24:  “we therefore define specifications as any patterns T that satisfy this inequality. In other words, specifications are those patterns whose specified complexity is strictly greater than 1.”  The inequality is χ > 1, which Dembski states is equivalent to 10 to the power 120, times φ, times P(T|H) being less than one half. ↩

    4. 3d.

      Page 25:  “is an evolutionary chance hypothesis that takes into account Darwinian and other material mechanisms”.  “The precise calculation of P(T|H) has yet to be done.” ↩ ↩

    5. 3e.

      Page 26:  “omits some crucial chance hypothesis that might be operating in the world and account for the event E in question.” ↩

  4. 4.

    Fitelson, Branden et al.  “How Not to Detect Design.  Critical Notice: William A. Dembski, The Design Inference.”  Philosophy of Science 66(3):472–488.  1999.  The scan prints two journal pages on each sheet.  Its text layer runs some words together;  spacing is restored in the quotations and nothing else is changed.

    1. 4a.

      Page 486:  “The Filter's treatment of Chance therefore applies only to agents who believe they have a complete list of the chance processes that might explain E.” ↩ ↩

    2. 4b.

      Page 486:  “Here Dembski is much too hard on Design.”  Page 487:  “What Paley did was compare a specific chance hypothesis and a specific design hypothesis without pretending that he thereby surveyed all possible chance hypotheses.” ↩ ↩

  5. 5.

    Dembski, William A.  No Free Lunch: Why Specified Complexity Cannot Be Purchased without Intelligence.  Rowman and Littlefield.  2002.  Dembski's sequel, aimed at biology.  It restates the criterion as complex specified information with a 500-bit threshold, argues that natural causes cannot generate it, answers the first critics of the 1998 book, and in chapter 5 computes the only probability Dembski has published for a biological structure, the bacterial flagellum.

    1. 5a.

      Page 14:  “Granted, this is a danger for the Explanatory Filter. But it is a danger endemic to all of scientific inquiry. Indeed, it is merely a restatement of the problem of induction”.  Answering John Wilkins and Wesley Elsberry by name. ↩ ↩

    2. 5b.

      Page 24:  “if things end up in the net that are not designed, the criterion is vitiated.” ↩

    3. 5c.

      Page 25:  “specification criterion successfully avoids false positives—in other words, whenever it attributes design, it does so correctly.”  The sentence begins on p. 24:  "I want to argue that the complexity-". ↩

    4. 5d.

      Page 26:  “Perhaps the best candidate is the natural nuclear reactors at the Oklo uranium mine in Gabon, West Africa.” ↩ ↩

    5. 5e.

      Page 27:  “The precise probabilities for such conditions have yet to be ascertained.”  “But suppose the Oklo reactors ended up satisfying this criterion after all. Would this vitiate the complexity-specification criterion? Not at all. At worst it would indicate that certain naturally occurring events or objects that we initially expected to involve no design actually do involve design.” ↩ ↩

    6. 5f.

      Page 101:  “Why should eliminating a chance hypothesis always require additional chance hypotheses that compete with it?” ↩

    7. 5g.

      Page 108:  “We infer design regularly and reliably without knowing characteristics of the designer or being able to assess what the designer is likely to do.” ↩

    8. 5h.

      Page 123:  “we might think we know the relevant chance hypotheses, but later discover that we missed a crucial one. In the one case a design inference could not even get going; in the other, it would be mistaken. But these are the risks of empirical inquiry, which of its nature is fallible.”  Endnote 80, answering Fitelson, Stephens and Sober. ↩ ↩ ↩

    9. 5i.

      Page 156:  “Accordingly, specified information of complexity greater than 500 bits cannot reasonably be attributed to chance.” ↩

    10. 5j.

      Page 159:  “The broad conclusion of the last section was therefore that natural causes are incapable of generating CSI.” ↩

    11. 5k.

      Page 289:  “Specification is never a problem. The irreducibly complex systems we consider, particularly those in biology, always satisfy independently given functional requirements (see section 3.7). For instance, in the case of the bacterial flagellum, humans developed outboard rotary motors well before they figured out that the flagellum was such a machine.” ↩

    12. 5l.

      Page 290:  “The challenge of irreducible complexity is to show that there are instances of biological complexity that must be attained all at once”. ↩

    13. 5m.

      Page 291:  “It follows that the probability of a discrete combinatorial object equals the product of these three probabilities”. ↩

    14. 5n.

      Page 293:  “Randomly picking 250 proteins and having them all fall among those 500 therefore has probability (500/4,289)^250, which has order of magnitude 10^-234 and falls considerably below the universal probability bound”.  The OCR loses both exponents.  They are recovered by recomputing;  see the note on the arithmetic. ↩ ↩

    15. 5o.

      Page 300:  “First, unlike the snowflake, which consists entirely of water molecules, the flagellum requires multiple diverse parts.” ↩

    16. 5p.

      Page 301:  “Then for a protein comprising 300 amino acid subunits, our quick and dirty approximation of its perturbation probability is”.  The figures that follow are 20^-30, or about 10^-39.  The OCR garbles them;  30 times log10 of 20 is 39.0. ↩

    17. 5q.

      Page 302:  “The big challenge is to firm up these numbers and make sure they do not cheat in anybody's favor.” ↩

    18. 5r.

      Page 377:  “Elsberry’s category of unknown causation amounts to a disingenuous display of ignorance—intelligence and only intelligence is known to generate specified complexity”.  Endnote 73.  Elsberry had proposed a version of the filter with a fourth outcome, "unknown causation." ↩ ↩

  6. 6.

    Matzke, N. J.  “Evolution in (Brownian) space: a model for the origin of the bacterial flagellum.”  TalkDesign.  2003.

    1. 6a.

      Section Abstract:  “The model consists of six major stages: export apparatus, secretion system, adhesion system, pilus, undirected motility, and taxis-enabled motility.” ↩ ↩

  7. 7.

    Gurevich, Yuri, and Grant Olney Passmore.  “Impugning Randomness, Convincingly.”  arXiv.  Dec 30, 2015.  The arXiv posting of a paper published as Studia Logica 100:193–222 (2012), doi 10.1007/s11225-012-9375-0.  Gurevich was at Microsoft Research.  Page numbers are the arXiv copy's.

    1. 7a.

      Page 9:  “the point is not that the focal event is specified in advance but that it is specified — possibly after the fact — without any knowledge about the outcome of the trial”. ↩ ↩

    2. 7b.

      Page 20:  “And he wants to impugn any chance whatsoever. That seems hopeless to us.” ↩

  8. 8.

    Ewert, Winston et al.  “On the Improbability of Algorithmic Specified Complexity.”  Proceedings of the 45th Southeastern Symposium on System Theory, 68–70.  2013.  Four-page conference paper by Ewert, Marks and Dembski that defines algorithmic specified complexity and proves that, under any given distribution, the chance of an outcome scoring α bits or more is at most 2 to the power −α.

    1. 8a.

      Page 69:  “we define algorithmic specified complexity (ASC)”.  The definition and the bound are displayed equations on this page. ↩ ↩

    2. 8b.

      Page 70:  “Given a high ASC object, we have evidence that the assumed probability distribution was incorrect.” ↩ ↩

  9. 9.

    Durston, Kirk K. et al.  “Measuring the functional sequence complexity of proteins.”  Theoretical Biology and Medical Modelling 4:47.  2007.

    1. 9a.

      Section Background:  “resulting in the equi-probability of all possible sequences or options”. ↩ ↩

    2. 9b.

      Section Background:  “The function could theoretically arise stochastically through mutational changes coupled with selection pressure, or through human experimenter involvement”. ↩

  10. 10.

    χ > 1 means 10120 × φ × P(T|H) < 1/2, so −log2 P(T|H) must exceed log2 10120 + log2 φ + 1.  log2 10120 = 120 × 3.3219 = 398.6.  The 2005 threshold is therefore about 400 bits plus log2 φ.  In 2002 the threshold was log2 10150 = 498.3, which Dembski rounded to 500. ↩

  11. 11.

    Dembski, William A.  “Reinstating the Explanatory Filter.”  Uncommon Descent.  Dec 10, 2008.

    1. 11a.

      Paragraph 1:  “I’m herewith reinstating it — it will appear, without reservation or hesitation, in all my future work on design detection.” ↩

  12. 12.

    Ewert, Winston et al.  “Algorithmic Specified Complexity in the Game of Life.”  IEEE Transactions on Systems, Man, and Cybernetics: Systems 45(4):584–594.  2015.

    1. 12a.

      Page 585:  “ASC provides evidence that a stochastic outcome modeled by the distribution, P(X), does not explain a given object.” ↩

  13. 13.

    Dembski, William A., and Winston Ewert.  The Design Inference: Eliminating Chance through Small Probabilities.  Discovery Institute Press.  2nd edition.  2023.  Only the publisher's front-matter file was obtained:  endorsements, contents with page numbers, and the foreword.  The body of the book was not read.

    1. 13a.

      Section Table of Contents:  “The Explanatory Filter”.  Listed as section 5.6, p. 244.  Chapter 6 is Specified Complexity (6.2 Description Length, 6.5 Frequentist Interpretation, 6.6 Bayesian Interpretation).  Chapter 7 is Evolutionary Biology (7.6 Doing the Calculation, p. 368). ↩

  14. 14.

    Dembski, William A.  “Specified Complexity Made Simple.”  billdembski.com.  Feb 26, 2024.  Dembski's own restatement of chapter 6 of the second edition.

    1. 14a.

      Section on specification:  “In fact, we define specifications as patterns with short descriptions.” ↩

    2. 14b.

      Section on the formal definition:  “is the probability of E with respect to some underlying relevant chance hypothesis.”  The formula SC(E) = I(E) – K(E) is printed in the sentence before. ↩

  15. 15.

    Dembski, William A., and Winston Ewert.  “An Argument from Ignorance?”  Science and Culture Today.  Nov 17, 2023.  An excerpt from the Introduction to the second edition of The Design Inference, released by the publisher.

    1. 15a.

      Paragraph 1:  “Note that the specification of these systems, as in their exhibiting the right sort of pattern for a design inference, was never in question.” ↩

    2. 15b.

      Paragraph 1:  “My approach in applying the design inference to biology was to piggyback on the work of design biologists such as Douglas Axe and Michael Behe.” ↩

    3. 15c.

      Section Misguided and Irrelevant:  “For Darwinists to refute ID, they merely needed to postulate unidentified, and perhaps forever unidentifiable, indirect Darwinian pathways”. ↩

    4. 15d.

      Section Dawkins’s “Laziness Challenge”:  “To seal the deal, however, the ID research community still needed something like the second law for biology.” ↩

  16. 16.

    Fisher, R. A.  “Has Mendel's Work Been Rediscovered?”  Annals of Science 1(2):115–137.  1936.  Read in the University of Adelaide reprint, which keeps the journal's page numbers in its running heads.

    1. 16a.

      Page 129:  “A total deviation of the magnitude observed, and in the right direction, is only to be expected once in 444 trials ; there is therefore here a serious discrepancy.” ↩

    2. 16b.

      Page 132:  “it remains a possibility among others that Mendel was deceived by some assistant who knew too well what was expected.” ↩

  17. 17.

    European Network of Forensic Science Institutes.  ENFSI Guideline for Evaluative Reporting in Forensic Science.  version 3.0 edition.  2015.

    1. 17a.

      Page 10:  “The findings should be evaluated given at least one pair of propositions”.  “If no alternative can be formulated, the value of the findings cannot be assessed.” ↩ ↩

  18. 18.

    Klimek, Peter et al.  “Statistical detection of systematic election irregularities.”  Proceedings of the National Academy of Sciences 109(41):16469–16473.  2012.  Read in the arXiv version.  Mirror:  arXiv

    1. 18a.

      Section Abstract:  “reported irregularities in recent Russian elections are indeed well explained by systematic ballot stuffing and develop a parametric model quantifying to which extent fraudulent mechanisms are present.” ↩

  19. 19.

    Bell Burnell, S. Jocelyn.  “Petit Four.”  Annals of the New York Academy of Sciences 302:685–689.  1977.  Read in the reprint in Cosmic Search 1(1), 1979, titled "Little Green Men, White Dwarfs or Pulsars?"  The reprint has no page numbers.

    1. 19a.

      About one third of the way through:  “If this were the case then the pulses should show Doppler shifts as the little green men on their planet orbited their sun.” ↩ ↩

    2. 19b.

      The paragraph on the conference before Christmas:  “We did not really believe that we had picked up signals from another civilization, but obviously the idea had crossed our minds and we had no proof that it was an entirely natural radio emission.” ↩

    3. 19c.

      After the second source is found:  “It was very unlikely that two lots of little green men would both choose the same, improbable frequency, and the same time, to try signalling to the same planet Earth.” ↩

  20. 20.

    Friedman, William F., and Elizebeth S. Friedman.  The Shakespearean Ciphers Examined.  Cambridge University Press.  1957.  Read on archive.org by paragraph search.  Not stored.  Pages are from the public leaf map, checked against the index entry "Feely, Joseph Martin ... 81-3."

    1. 20a.

      Page 21:  “The mathematical theory of probability can be applied, and the chances calculated exactly. If the cryptanalyst finds a certain key and (on the basis of the way it is built up) he calculates that the chances of its appearing by accident are one in one thousand million, his confidence in the solution will be more than justified.” ↩ ↩ ↩

  21. 21.

    Warren, S. Hazzledine.  “On the Origin of "Eolithic" Flints by Natural Causes, especially by the Foundering of Drifts.”  Journal of the Anthropological Institute of Great Britain and Ireland 35:337–364.  1905.

    1. 21a.

      Page 338:  “so much reliance has been placed by many of our geologists and archeologists”.  “It is argued from this that they must be due to intelligent design on the part of man. To this I venture to give a flat non sequitur.”  “can we, apart from this, prove that this chipping must have been done by man, and could not have been done by nature?” ↩ ↩

  22. 22.

    Eklund, Sigvard.  “Oklo: A Nuclear Reactor 1800 Million Years Ago.”  IAEA Bulletin 17(3):2–4.  1975.  Eklund was Director General of the International Atomic Energy Agency.

    1. 22a.

      Page 2:  “This is done primarily to ensure that the contained uranium really is of natural origin and does not come from a plant where some fraction of the uranium-235 has already been used.” ↩

    2. 22b.

      Page 2:  “As early as the beginning of August 1972 the idea was put forward that the depletion in uranium-235 had been caused by a chain reaction a long time ago”.  Page 3:  “which can be identified mass spectrometrically on the basis of their isotopic composition.” ↩

  23. 23.

    Sober, Elliott.  Evidence and Evolution: The Logic Behind the Science.  Cambridge University Press.  2008.  Not obtained in full.  Read as Google Books search snippets, which print a page label with each snippet.  The labels were not checked against a printed copy.  Dembski is indexed at pp. 51 and 168 only.

    1. 23a.

      Page 51:  “Probabilistic modus tollens is an incorrect form of inference”. ↩

  24. 24.

    Sober, Elliott.  “The Design Argument.”  God and Design: The Teleological Argument and Modern Science, 27–54.  Routledge.  Edited by Neil A. Manson.  2003.  A revised version appeared in The Blackwell Guide to the Philosophy of Religion, 2004.  Read from a scan of the 2003 printing, folios at the foot of each page.

    1. 24a.

      Page 38:  “The problem is that the design hypothesis confers a probability on the observation only when it is supplemented with further assumptions about what the designer's goals and abilities would be if he existed.” ↩

    2. 24b.

      Page 39:  “they often have objects in their collections that they know are artifacts, although they have no idea what the makers of these artifacts had in mind.” ↩

    3. 24c.

      Page 40:  “The SETI engineers search for this signal, not because it is "complex" or fulfills some a priori criterion that would make it a "sign of intelligence," but simply because they think they know what sorts of mechanisms are needed to produce it.” ↩

  25. 25.

    Enriquez, J. Emilio et al.  “The Breakthrough Listen Search for Intelligent Life: 1.1–1.9 GHz Observations of 692 Nearby Stars.”  The Astrophysical Journal 849:104.  2017.  Radio search with the Green Bank Telescope of 692 nearby stars for narrowband signals drifting in frequency, as a transmitter on a moving planet would produce.  Every candidate was consistent with human radio interference.  The paper states in its own words why SETI looks for narrowband signals.  Mirror:  arXiv

    1. 25a.

      Section 3, Data analysis:  “This type of signal is of particular interest in traditional SETI projects because it is too narrow to arise naturally from known natural astrophysical sources, and represents a power-efficient method of transmitting a beacon signal out to great distances.” ↩

  26. 26.

    International Academy of Astronautics.  “Declaration of Principles Concerning Activities Following the Detection of Extraterrestrial Intelligence.”  IAA SETI Permanent Committee.  1989.  The protocol SETI researchers agreed to follow after a possible detection, adopted by the International Academy of Astronautics.  Its first principle sets out how a discoverer should test a signal before announcing it.

    1. 26a.

      Paragraph principle 1:  “should seek to verify that the most plausible explanation for the evidence is the existence of extraterrestrial intelligence rather than some other natural phenomenon or anthropogenic phenomenon before making any public announcement. If the evidence cannot be confirmed as indicating the existence of extraterrestrial intelligence, the discoverer may disseminate the information as appropriate to the discovery of any unknown phenomenon.” ↩

  27. 27.

    500 divided by 4,289 is 0.1166, and its base-10 logarithm is −0.9334.  Multiplied by 250 that is −233.3, so the probability is about 5 × 10−234, which Dembski rounds to 10−234.  For Caputo, Olofsson counts 42 sequences with at least 40 Democrats out of 241, or about 1.9 × 10−11. ↩

  28. 28.

    Olofsson, Peter.  “Intelligent design and mathematical statistics: a troubled alliance.”  Biology and Philosophy 23(4):545–553.  2008.

    1. 28a.

      Page 550:  “My first objection is that Dembski considers only the outcome, not the rejection region.”  “how could he possibly address what it means to have”. ↩

    2. 28b.

      Page 550:  “The second problem with Dembski’s application is that he only tests one particular chance hypothesis and takes its rejection as evidence that all chance hypotheses can be ruled out. In particular, the chance hypothesis he considers is based on a uniform distribution or, in daily language, that the flagellum is assembled at random.” ↩

  29. 29.

    Häggström, Olle.  “Intelligent Design and the NFL Theorems.”  author's preprint.  Jun 2, 2006.  Published as Biology and Philosophy 22:217–230 (2007), doi 10.1007/s10539-006-9040-z.  Page numbers here are the preprint's.

    1. 29a.

      Page 12, footnote 10:  “such uniform distributions have no privileged status over other models as realistic descriptions of what the laws of nature produce”. ↩

  30. 30.

    Liu, Renyi, and Howard Ochman.  “Stepwise formation of the bacterial flagellar system.”  Proceedings of the National Academy of Sciences 104(17):7116–7121.  2007.  Mirror:  PMC

    1. 30a.

      Section Abstract:  “many of these core genes show sequence similarity only to other flagellar core genes, indicating that they were derived from one another”. ↩

  31. 31.

    Rosenhouse, Jason.  “New Edition of Dembski's Design Inference.”  The Panda's Thumb.  Nov 19, 2023.  A blog post by a mathematician at James Madison University, written from the published excerpt before he had the book.

    1. 31a.

      Paragraph 3:  “This means his argument is completely parasitic on Michael Behe’s claims about IC systems”. ↩

  32. 32.

    Felsenstein, Joe.  “Dembski and Ewert wish upon a star.”  The Panda's Thumb.  Nov 10, 2024.  A blog post reviewing chapter 7 of the second edition, section by section, with the book's page ranges.

    1. 32a.

      Section Doing the Calculation (pages 368-378):  “Calculates the probability of a “biological system”, using protein folds as one example, invoking Douglas Axe’s experiments”.  “Dembski and Ewert do not try to quantitate the length of the “description” needed.” ↩

    2. 32b.

      Section What have Dembski and Ewert accomplished?:  “There is simply no discussion of how to compute a tail probability in some relevant distribution.” ↩

  33. 33.

    Dembski, William A.  The Design Revolution: Answering the Toughest Questions about Intelligent Design.  InterVarsity Press.  2004.  A book of short chapters, each answering one objection to intelligent design.  Chapter 14, "Assertibility," argues that a design claim can be justified on present evidence when the probability under evolutionary hypotheses cannot be computed, and that the flagellum meets that standard.

    1. 33a.

      Page 113:  “science must form its conclusions on the basis of available evidence, not on the possibility of future evidence. If evolutionary biologists can discover or construct detailed, testable, indirect Darwinian pathways that account for the emergence of irreducibly and minimally complex biological systems like the bacterial flagellum, then more power to them — intelligent design will quickly pass into oblivion.” ↩

    2. 33b.

      Page 114:  “Configuration spaces like this are perfectly homogeneous, with one character string geometrically interchangeable with the next. The geometry therefore precludes any underlying mechanisms from distinguishing or preferring some character strings over others.”  “Michael Polanyi made such arguments for biological design in the 1960s. Stephen Meyer has updated them for the present.”  “It is therefore incumbent on the scientific community to admit, at least provisionally, that the bacterial flagellum is designed.”  The spaces meant are "all possible character sequences from a fixed alphabet," which Dembski says model "polymers like DNA, RNA and proteins." ↩

    3. 33c.

      Page 115:  “That the bacterial flagellum exhibits specified complexity is highly assertible — this despite the logical impossibility of ruling out the infinity of possible indirect Darwinian pathways that might give rise to it.” ↩

  34. 34.

    Kepler, Johannes.  Kepler's Somnium: The Dream, or Posthumous Work on Lunar Astronomy.  University of Wisconsin Press.  Translated by Edward Rosen.  1967.  The passage is the Geographical Appendix, a letter to Paul Guldin written after 17 July 1623 and printed in 1634.  Read on archive.org by paragraph search.  Not stored.  Pages are from the public leaf map only;  a second check against the running heads was not possible.

    1. 34a.

      Page 151:  “If you direct your mind to”.  The sentence continues "the towns on the moon, I shall prove to you that I see them," with an editor's mark between.  The stake and ropes follow on the same page. ↩

    2. 34b.

      Page 152:  “But these are playful remarks”. ↩

  35. 35.

    Proffitt, Tomos et al.  “Wild macaques challenge the origin of intentional tool production.”  Science Advances 9(10):eade8159.  2023.  Mirror:  PMC

    1. 35a.

      Section Abstract:  “In the absence of behavioral observations, the assemblage produced by monkeys would likely be identified as anthropogenic in origin and interpreted as evidence of intentional tool production.” ↩

  36. 36.

    Libbrecht, Kenneth G.  “The physics of snow crystals.”  Reports on Progress in Physics 68:855–895.  2005.

    1. 36a.

      Page 862:  “But of course all six arms experience the same changing conditions as they grow.” ↩

  37. 37.

    Elsberry, Wesley, and Jeffrey Shallit.  “Information theory, evolutionary computation, and Dembski's "complex specified information".”  authors' manuscript.  Nov 16, 2003.  The authors' full 54-page version.  The journal's referees cut it to 34 pages, published as Synthese 178:237–270 (2011), doi 10.1007/s11229-009-9542-8.  Page numbers here are the manuscript's.  The published pagination was not seen.

    1. 37a.

      Page 7:  “any observed event with a sufficiently complicated or obscure causal history could mistakenly be assigned to design”. ↩

    2. 37b.

      Page 10:  “By its arbitrary demand on video-camera-certainty, Dembskian induction rules out considering cases where the underlying causal story cannot be known with this form of certainty”. ↩

  38. 38.

    Wilkins, John S., and Wesley R. Elsberry.  “The advantages of theft over toil: the design inference and arguing from ignorance.”  Biology and Philosophy 16:711–724.  2001.  Paper by the philosopher of biology John Wilkins and the biologist Wesley Elsberry.  It argues that the filter treats ignorance of a natural cause as evidence of design, and that real design inferences keep a "don't know" outcome the filter lacks.

    1. 38a.

      About halfway, after the discussion of reliability:  “What the filter lacks that real-world design inferences already have is a "Don't know" decision.” ↩

  39. 39.

    Dembski, William A.  “Design by Elimination vs. Design by Comparison.”  billdembski.com.  Sep 2005.  The author's posted copy, 12 pages, of chapter 33 of The Design Revolution (InterVarsity Press, 2004).  Page numbers are the posted copy's.

    1. 39a.

      Page 12:  “The Bayesian approach to statistical rationality is parasitic on the Fisherian approach and can properly adjudicate only among hypotheses that the Fisherian approach has thus far failed to eliminate.” ↩

  40. 40.

    Wolpert, David.  “William Dembski's treatment of the No Free Lunch theorems is written in jello.”  Talk Reason.  2002.  Also Mathematical Reviews MR1884094.  Wolpert is co-author of the No Free Lunch theorems.  Mirror:  Archive.is

    1. 40a.

      Paragraph 4:  “his arguments are fatally informal and imprecise.” ↩

  41. 41.

    Felsenstein, Joe.  “Has Natural Selection Been Refuted? The Arguments of William Dembski.”  Reports of the National Center for Science Education 27(3–4):20–26.  2007.

    1. 41a.

      Section Dembski's proof:  “He does this by changing the specification”. ↩

  42. 42.

    Devine, Sean.  “An Algorithmic Information Theory Challenge to Intelligent Design.”  Zygon 49(1):42–65.  2014.

    1. 42a.

      Page 42, abstract:  “Dembski’s decision process is flawed, as natural explanations are eliminated before chance.” ↩

  43. 43.

    Durston, Kirk.  “Why this scientist believes intelligent design was required for biological life.”  kirkdurston.com.  Jul 22, 2019.

    1. 43a.

      Paragraph the stated hypothesis:  “A unique property of intelligent minds is the ability to produce statistically significant levels of functional information as defined in the literature.” ↩

  44. 44.

    Felsenstein, Joe.  “Dembski and Ewert dump functional information.”  The Panda's Thumb.  Jun 12, 2024.  A blog post by a population geneticist at the University of Washington, reviewing chapters 1 to 6 of the second edition.

    1. 44a.

      Paragraph 3:  “Dembski and Ewert have moved to definition 3, resulting in a specification that no longer has any connection to Functional Information.” ↩

  45. 45.

    Behe, Michael J., and David W. Snoke.  “Simulating Evolution by Gene Duplication of Protein Features That Require Multiple Amino Acid Residues.”  Protein Science 13(10):2651–2664.  2004.  Population-genetic model by the biochemist Michael Behe and the physicist David Snoke of a duplicated gene that must gain a feature needing two or more amino acids, where the single steps do not help.  It is a probability under a named mechanism, the kind of number the filter needs, for one pathway that the authors say may not be the only one.  Mirror:  PMC

    1. 45a.

      Page 2651:  “We conclude that, in general, to be fixed in 108 generations, the production of novel protein features that require the participation of two or more amino acid residues simply by multiple point mutations in duplicated genes would entail population sizes of no less than 109.”  The exponents are superscripts in the paper. ↩

  46. 46.

    Lynch, Michael.  “Simple evolutionary pathways to complex proteins.”  Protein Science 14(9):2217–2225.  2005.  Reply to Behe and Snoke by a population geneticist.  Lynch reran the problem with neutral intermediate steps and more than one usable site, and found that such features can arise in a million years or less in populations of moderate size.  Mirror:  PMC

    1. 46a.

      Page 2217:  “can evolve on time scales of a million years (or much less) in populations of only moderate size.” ↩

  47. 47.

    Durrett, Rick, and Deena Schmidt.  “Waiting for two mutations:  with applications to regulatory sequence evolution and the limits of Darwinian evolution.”  Genetics 180(3):1501–1509.  Nov 2008.  Mathematical population genetics by two Cornell mathematicians, written in part to answer Michael Behe.  It computes, with a theorem checked by simulation, how long a population waits for a named pair of mutations when the second is favoured by selection.  Mirror:  PMC

    1. 47a.

      Page 1501:  “Consistent with recent experimental observations for Drosophila, we find that a few million years is sufficient, but for humans with a much smaller effective population size, this type of change would take >100 million years.” ↩

    2. 47b.

      Page 1507:  “If one can search for the new target sequence in 104 – 106 bp, then there are many more chances.”  The target is a binding site for a gene-regulating protein, which may sit anywhere in a regulatory region thousands to a million DNA letters long. ↩

  48. 48.

    Gauger, Ann K., and Douglas D. Axe.  “The evolutionary accessibility of new enzyme functions: a case study from the biotin pathway.”  BIO-Complexity 2011(1):1–17.  2011.  Mutagenesis study from the Biologic Institute, an intelligent-design laboratory.  The authors tried to convert the E. coli enzyme Kbl into its structural relative BioF, inferred that at least seven base changes are needed, and fed that into a population model with duplication, mutation and selection.

    1. 48a.

      Page 12:  “we estimate that some 1030 or more generations would elapse before a bioF-like innovation that is paralogous to kbl could become established”.  “we conclude not that paralogous evolution absolutely cannot have accomplished a functional jump like the one examined here, but rather that there is now a scientific case for doubting this particular jump to be evolutionarily feasible.”  The exponent is a superscript in the paper. ↩

  49. 49.

    Sanford, John et al.  “The waiting time problem in a model hominin population.”  Theoretical Biology and Medical Modelling 12:18.  Sep 17, 2015.  Computer simulation by a team led by the plant geneticist John Sanford, a young-earth creationist, using their program Mendel's Accountant, of how long a population of at least 10,000 human ancestors with strong selection waits for a named short DNA string to arise and become fixed.  The target is one specified string, so the result applies to a change with exactly one acceptable outcome.

    1. 49a.

      Section Abstract, Results:  “To establish a string of two nucleotides required on average 84 million years. To establish a string of five nucleotides required on average 2 billion years.” ↩

  50. 50.

    Orgel, Leslie E.  The Origins of Life: Molecules and Natural Selection.  John Wiley and Sons.  1973.  Read on archive.org by paragraph search.  Not stored.  The page is from the public leaf map and agrees with the page Dembski cites.

    1. 50a.

      Page 189:  “living organisms are distinguished by their specified complexity.” ↩

  51. 51.

    Dawkins, Richard.  The Blind Watchmaker.  W. W. Norton.  1986.  Read on archive.org by paragraph search.  Not stored.  Printed page = leaf minus 22, from the public leaf map.

    1. 51a.

      Page 9:  “complicated things have some quality, specifiable in advance, that is highly unlikely to have been acquired by random chance alone.” ↩